Skip to content

Self-hosted, Self-Improving, Federated Everything App — dev management, agent orchestration, persistent mind, digital twin, second brain, and creative pipeline from a single dashboard over Tailscale to port the user into the OS

fix: distinguish provider defaults from effective shot stage timeouts (#13204)

[._.]/ Adam Eivyf40ec02d22a9
diff --git a/docs/features/fableloom.md b/docs/features/fableloom.md
index 0a37638eaf..28ca8128b1 100644
--- a/docs/features/fableloom.md
+++ b/docs/features/fableloom.md
@@ -426,6 +426,6 @@ The shot planner starts a background run with `POST /api/fableloom/:id/episodes/
`GET .../shots/status` reattaches to its latest state; `POST .../shots/cancel` stops after the active provider call.
The UI receives `fableloom:shots:run` snapshots and reloads status on reconnect. The legacy `POST .../shots/plan` remains synchronous for existing callers.
-Draft and review run on one source scene at a time with the episode's compact graph digest. A final text consistency review receives the complete shot plan and graph, without the complete teleplay. Provider timeout settings remain authoritative; large inputs show a warning. Stage failures name the stage, round, scene and provider.
+Draft and review run on one source scene at a time with the episode's compact graph digest. A final text consistency review receives the complete shot plan and graph, without the complete teleplay. Existing stage and provider timeout settings stay in effect; large inputs show a reminder to check the effective timeout. Stage failures name the stage, round, scene and provider.
Each draft and review is saved before the next call. After a restart the run is reported failed with those steps preserved; no AI work starts at boot. Resume sends `{ "resume": true }` to the start endpoint, retaining the original route, direction and apply choice, and reuses completed steps. Changed episodes require a fresh plan. New previews explicitly start a fresh run. Applying still requires a complete, validated, reviewed plan matching the episode revision; cancellation prevents application after an in-flight call.
diff --git a/server/services/fableLoom/shots.js b/server/services/fableLoom/shots.js
index fa2298d645..4334d972ab 100644
--- a/server/services/fableLoom/shots.js
+++ b/server/services/fableLoom/shots.js
@@ -111,9 +111,9 @@ export async function runEpisodeShotAutopilot(id, episodeId, {
const selectedProvider = resolved.provider;
const inputCharacters = Object.values(variables).reduce((total, value) => total + String(value).length, 0);
checkpoint.current = { stage, round, sceneId, providerId: selectedProvider.id,
- inputCharacters, timeoutMs: selectedProvider.timeout,
- warning: inputCharacters > 12000 && (selectedProvider.timeout || 300000) <= 300000
- ? 'Large shot-planning input; this provider timeout may be too short. Completed steps will be preserved on failure.' : null };
+ inputCharacters, providerTimeoutMs: selectedProvider.timeout,
+ warning: inputCharacters > 12000
+ ? 'Large shot-planning input; check the effective stage/provider timeout. Completed steps will be preserved on failure.' : null };
await persist();
return runStagedLLM(stage, variables, { ...route, source: stage,